Recently, in the Hong Kong stock market, Zhipu and MiniMax, often referred to as the "dynamic duo of large models," have made quite a splash.

On July 9, Zhipu took the lead by announcing the initiation of a placement worth approximately HK$31.4 billion, setting a new benchmark for the largest single placement funding scale among Hong Kong-listed technology companies for the year. Immediately thereafter, on July 10, MiniMax announced the completion of a new round of financing totaling HK$16 billion.

The timing of these two financing events coincided with the window period during which the initial lock-up shares of both companies were released. However, the outcomes were starkly different. Zhipu's stock price soared by 15% on the day, while MiniMax's stock price plummeted by 17%.

The billion-dollar-level financings and the wave of lock-up expirations have put the capital reserves of both companies to the test, while also representing a profound shift in the valuation logic for domestic large models by the capital market.

Chip Structure and Liquidity Pricing: The Truth Behind the "Fire and Ice" of the Lock-up Release Wave

Lock-up release refers to the right of previously restricted shares, which were temporarily non-tradable, to be traded on the market.

The event of lock-up release does not inherently cause stock prices to rise or fall. Whether stock prices rise or fall depends on whether the relevant shareholders holding the shares choose to "sell" or "hold" after the lock-up release, as well as the volume of share transactions.

Zhipu's high-valuation placement was a meticulously planned capital maneuver. On July 8, following its market debut, Zhipu experienced the release of its initial lock-up shares, totaling approximately 25.6816 million shares, accounting for 5.76% of its total share capital. The entities subject to this lock-up release included 11 investors with state-owned backgrounds and attributes of large financial institutions, such as JSC International, Taikang Life Insurance, Guangfa Fund, Shanghai Gaoyi Asset Management, and WT Asset Management.

These types of shareholders do not aim for short-term arbitrage and have a strong intention to hold onto their shares. On the eve of the lock-up release, nearly 70% of the relevant shareholders publicly stated their intention to hold the shares for the long term. Therefore, Zhipu's lock-up release did not trigger a large-scale selling pressure. Even if all the shares were sold, the total volume of the released shares was relatively small, creating a bullish expectation in the market, which is why Zhipu's stock price rose.

Zhipu's management capitalized on the situation by completing a HK$31.4 billion new share placement in one go at an approximately 13% discount, achieving a remarkable financing feat when market valuation and equity liquidity were high, thereby preparing financial reserves for subsequent technological competitions.

MiniMax's financing move, on the other hand, resembled a defensive battle to stabilize morale. On July 9, MiniMax had approximately 150 million shares released from lock-up, accounting for nearly half (44.85%) of its total share capital. Among the entities subject to this lock-up release, in addition to strategic shareholders like Alibaba and miHoYo, there were also venture capital firms such as Hillhouse Capital, Sequoia Capital, and IDG Capital.

Venture capital firms are typical financial investors with clear holding periods and return rate assessments. Their basic business model is "invest - exit - profit." They are likely to gradually sell their shares after the lock-up release. Coupled with the relatively high proportion of MiniMax's shares being released, the market might struggle to absorb them in the short term, leading to a bearish expectation and causing MiniMax's stock price to slide.

Therefore, on July 10, MiniMax announced its intention to place 35.6 million new shares and issue HK$6.5 billion in zero-coupon convertible bonds, with an expected total financing amount of approximately HK$16 billion, aiming to stabilize market sentiment.

The differentiated performances of the two companies in their financing approaches are also worth scrutinizing. Zhipu directly placed new shares at a 13% discount, rapidly absorbing institutional funds with the aim of fully raising capital when stock prices were high to prepare ammunition for its next development plans.

MiniMax adopted a combination of new share placement and zero-coupon convertible bonds, with the new share placement used to immediately supplement cash flow and the zero-coupon convertible bonds conveying confidence in long-term value to the market through their premium conversion prices.

There is no superiority or inferiority between these two financing methods; both are pragmatic "optimal solutions." Zhipu leveraged the stable foundation of state-owned shareholders and industrial capital to create market conditions for high-valuation financing, while MiniMax, due to its high proportion of financial investors, needed to balance financing efficiency and market confidence amid stock price fluctuations.

Divergence in Business Paths: Liu Debing's "B2B Moat" vs. Yan Junjie's "C-end Globalization"

In terms of scale, Zhipu currently has a total market capitalization of approximately HK$800 billion (as of the time of writing), while MiniMax has a total market capitalization of about HK$90 billion (as of the time of writing). The two companies are in different leagues and have varying levels of risk resistance, which is also one of the contributing factors to this lock-up release turmoil.

The short-term fluctuations in stock prices are superficial. Through this lock-up release turmoil, the outside world can discern a fact—the market's valuation logic for the two companies is already on different tracks. The reason for their divergence lies in their differing business models.

Zhipu follows a B2B route, acting as a "shovel seller" in the AI era. It provides foundational large model capabilities to government enterprises, developers, and enterprise clients, charging based on token usage or project-based fees, a typical MaaS (Model as a Service) model. Due to the strong payment capacity and high renewal rates of government enterprise clients, Zhipu's revenue stability is quite remarkable.

In 2025, Zhipu's annual revenue reached RMB 724 million, with a year-on-year increase of 132% and an overall gross margin of 41%. In times of rising market risk aversion, companies with such stable income expectations are more favored by institutional investors.

MiniMax, on the other hand, commercializes around C-end users and AI applications. With C-end products like Talkie and Xingye, it has amassed a large monthly active user base. Its C-end model has the advantages of a relatively high user scale ceiling and ease of dissemination on social media, but it also has the disadvantages of relatively low user payment willingness and the need for continuous marketing expenses to guide users, resulting in heavier profitability pressures.

In 2025, MiniMax's annual revenue was approximately RMB 560 million, with a year-on-year increase of 158.9%. However, its overall gross margin was 25.4%, and its overall operating condition still lags behind that of Zhipu.

Different commercialization paths have also formed different industry ecological niches. Zhipu focuses on productivity scenarios and has excelled in AI programming business development. Its independently developed new-generation open-source large language model, GLM-5.2, has a comprehensive reasoning cost that is approximately one-sixth of that of overseas top models.

MiniMax focuses on social entertainment scenarios, creating emotional value through AI virtual character interactions, with typical product representatives being Talkie overseas and Xingye domestically. In cultural creation scenarios, MiniMax's Hailuo AI has generated over 600 million videos, also gaining a certain market presence.

Regarding overseas expansion, the two companies have different considerations. MiniMax's revenue relies on user base, making overseas expansion a necessity. Last year, approximately 70% of its revenue came from international markets, with business coverage in over 200 countries and regions, and its overseas commercialization is in a mature stage.

Zhipu's operating revenue relies more on the domestic market, so its pace of overseas expansion is relatively slow, and its overseas commercialization is still in the initial exploration stage.

Overall, Zhipu relies on the certain income from mature businesses, while MiniMax bets on the incremental market of AI for C-end users. If managed well, both can achieve profitability. However, the current AI industry has entered an adjustment period, where certainty holds more value in the capital market.

Yan Junjie's "All-in" Move and Liu Debing's "Steady Blood Transfusion"

The business model is the foundation for a company's growth, and the decision-making direction of a company during critical periods is largely determined by the personality traits of its leader.

Yan Junjie's "zero-salary" declaration is a vivid emotional anchor point. At a critical moment when the stock price was plummeting and there was constant noise in the market, MiniMax's founder, Yan Junjie, issued a letter to all employees, stating that he would not receive any salary until AGI (Artificial General Intelligence) was achieved. He also offered 5% of his shares to incentivize the team and support open-source initiatives.

Yan Junjie's actions were primarily aimed at stabilizing internal morale. As the large model industry enters the deep waters, the iteration speed is determined not only by computing power but also by the stability of the core team under high-pressure environments. By offering shares to incentivize the team, he reinforced the internal organization's resilience, ensuring the retention of talent and computing power from a material perspective. This sacrifice of personal interests also, to a certain extent, offset market sentiment.

Liu Debing, the chairman of Zhipu, possesses a pragmatic and progressive mindset. When the company's stock price surged significantly after the lock-up release and market sentiment was high, he did not dwell on the short-term figures on the books and decisively launched a massive new share placement plan. The funds raised will be primarily used for the iteration of the GLM foundational large model, the development of its own AI chips, and the expansion of domestic computing power clusters.

This pragmatic and slightly aggressive decision-making style is actually buying time for long-term research and development. Currently, all companies are vying for high-computing-power AI chips. Due to the inherent bottleneck in the supply of high-performance computing hardware, stockpiling funds as early as possible allows for the early locking of computing power resources, thereby widening the gap in model iteration.

Given their different situations, there is no right or wrong between the two response methods. Yan Junjie exchanged part of his personal interests for internal stability to maintain the core talent team during the company's downturn, while Liu Debing used short-term equity dilution to acquire cash reserves to solidify his competitive edge when the market is on the rise.

Ultimately, both adhere to a long-term perspective, striving to create room for trial and error in the company's development.

Final Conjecture: The "Matthew Effect" in the Computing Power Arms Race

From a longer-term perspective, the differences arising from the equity circulation lock-up releases of Zhipu and MiniMax are merely a microcosm of the industry.

The true industry backdrop is that the large model sector, particularly the domestic large model industry, has reached the deep waters, and the Matthew effect is accelerating. Several development trends have become evident.

First, capital reserves are a matter of life and death. The large model industry is a typical capital-intensive sector. Zhipu spent over 93% of its IPO financing proceeds within half a year of listing, and MiniMax's cash flow is also under intense strain.

Through the HK$31.4 billion and HK$16 billion financings, the two companies will have sufficient development funds for the next two to three years. They can also pre-lock scarce high-computing-power chips, ensuring the rhythm of model iteration and continuously expanding their leading advantages as top enterprises.

Second, the valuation logic has shifted from "storytelling" to "gross margin competition." In the past, the capital market would assign valuation premiums to vague concepts due to the scarcity of domestic large models. Now, investors are more rational and have started to scrutinize the core operating indicators of large model companies.

In the upcoming financial reports, Zhipu needs to demonstrate the scalability of its B2B business, while MiniMax needs to show signs of improvement in C-end customer revenue. Companies that fail to meet commercialization goals will face sustained downward adjustments in their stock prices.

Finally, both companies are advancing their open-source ecosystems and dual-capital-market layouts simultaneously. Zhipu has opened up its model weights to attract developers, while MiniMax plans to open-source its large model parameters to build technological influence. In essence, both are relying on open-source means to establish ecological barriers.

Moreover, both companies are pushing forward their domestic listing plans to achieve dual-market financing channels in Hong Kong and on the A-share market. Zhipu has initiated coaching for its Science and Technology Innovation Board IPO, while MiniMax has also started researching RMB share issuance. Both companies intend to connect with domestic industrial capital and policy resources to further solidify their long-term competitiveness.

Regardless, in the current domestic large model sector, although the competition has not yet reached its final stage, everyone can sense that a shakeout is imminent. Only participants with sufficient funds, strong technology, and clear commercialization paths will be able to persist in the long-term computing power competition.

As for which of Zhipu and MiniMax can go further, we will just have to wait and see.